NVIDIA has introduced a comprehensive suite of open-source robotics AI models and simulation tools designed to advance humanoid robot research and development. The company’s new releases, including the Newton Physics Engine, Isaac GR00T foundation model, and Cosmos simulation libraries, aim to help developers build more adaptable, efficient, and intelligent robotic systems. By combining GPU-accelerated simulation, scalable AI infrastructure, and open foundation models, NVIDIA is positioning itself as a central player in shaping the next phase of industrial and service robotics.
The Newton Physics Engine stands out as one of the most significant innovations in this release. It is an open-source, GPU-accelerated physics simulation engine developed under the Linux Foundation in collaboration with Google DeepMind and Disney Research. The tool allows developers to create realistic simulations that replicate complex environments and physical interactions, including walking on uneven surfaces, manipulating delicate objects, and balancing intricate robotic movements. According to NVIDIA, Newton is now available in Isaac Lab, the company’s platform for robot learning and simulation. Early adopters of this technology include ETH Zurich, Technical University of Munich, Peking University, and robotics manufacturer Lightwheel. Rev Lebaredian, NVIDIA’s Vice President of Omniverse and simulation technology, highlighted that humanoids represent the next frontier of physical AI, requiring robots that can reason, adapt, and operate safely in unpredictable environments.
Complementing Newton, NVIDIA has introduced the Isaac GR00T N1.6 foundation model, a system designed to serve as the cognitive core for robotic intelligence. GR00T integrates Cosmos Reason, a vision-language-action model that helps robots interpret vague or abstract human instructions and convert them into logical, step-by-step action plans. Cosmos Reason has already achieved top rankings on Hugging Face’s Physical Reasoning Leaderboard, signaling its strength in contextual understanding and decision-making. In addition, NVIDIA has enhanced its Cosmos world foundation models, including Cosmos Predict 2.5 and Cosmos Transfer 2.5, which can generate large-scale training data with improved accuracy. These updates allow robots to learn from multiple camera perspectives and simulate realistic interactions, enhancing their ability to perform complex tasks such as using tools or handling multiple objects simultaneously.
To support these advancements, NVIDIA has also launched new hardware infrastructure optimized for robotics and AI workloads. The GB200 NVL72 system, equipped with 72 Blackwell GPUs, provides high-performance computing capabilities for cloud-scale robotics training. RTX PRO Servers are now being deployed for unified AI simulation, training, and inference workflows, while Jetson Thor, an edge AI platform, allows robots to process multi-AI tasks directly on-device. Companies like Figure AI, Skild AI, and Unitree are already leveraging Jetson Thor to integrate advanced AI capabilities into their robotic systems.
NVIDIA’s technologies are increasingly shaping academic and industrial research efforts. Nearly half of the papers presented at the recent Conference on Robot Learning (CoRL) cited NVIDIA’s frameworks, reflecting their growing influence in the field. Research institutions such as Stanford University, Carnegie Mellon University, and the National University of Singapore are already utilizing NVIDIA’s GPU architecture and simulation frameworks for advancing physical AI. With Newton, GR00T, and Cosmos now openly available, NVIDIA is strengthening its role in enabling a new generation of humanoid and industrial robots capable of reasoning, learning, and acting with greater precision and autonomy.
Follow the SPIN IDG WhatsApp Channel for updates across the Smart Pakistan Insights Network covering all of Pakistan’s technology ecosystem.
